Table 1 Trade and growth
a
1 2
3 4
5 0.869
0.861 0.910
0.806 0.855
Dy
it−r
21.1 27.3
26.8 25.3
33.3 Dik
1.295 1.496
1.516 1.181
1.592 8.1
11.4 11.9
7.5 17.9
Dedu −
0.016 −
0.014 −
0.027 −
0.008 −
0.015 −
1.3 −
1.7 −
2.7 −
0.8 −
2.4 –
0.225 Dspec.
– –
0.122 –
– 0.156
Ldiss. –
0.067 –
– 5.4
– 1.4
DACelec –
– –
0.009 0.029
– –
– 2.8
5.9 −
0.402 −
0.162 M
2
− 0.959
− 0.048
0.036 Sargan test
20.7 18 27.8 23
25.3 23 28.3 23
29.1 27 No. of obs.
117 117
117 117
117
a
Dependant variable, y
it
− y
it−r
. T-statistics are given under the value of the estimated coefficients.
Differencing allows to drop the individual effects. The above equation will be estimated in what follows, with different specifications for the explanatory variables
W
it
. Data sources are described in Appendix A.
4. The results
The dependant variable is y
it
− y
it − r
, the first difference of the logarithm of real GDP per worker, r = 5. Following the ‘augmented Solow’ model, the most immedi-
ate variables considered for inclusion as explanatory variables in a growth regres- sion are education and investment. The first education variable to be considered in
this paper is the average number of years of secondary schooling of the total population. The investment variable is the ratio of investment to capital stock, i.e.
the renewal rate of the capital stock. The inclusion of this term in the regressions can be justified on several grounds. First, it comes as a near-direct consequence of
a specification of a production function with capital included as a production factor. Second, it could be interpreted as a consequence of a technical progress
function in the spirit of Kaldor 1957 or Kaldor and Mirlees 1962. When technological progress is embodied in equipment, the renewal rate of capital
influences positively the rate of productivity growth. More generally, the investment rate or its variation can be interpreted as a proxy for various factors influencing
technical progress, either learning by doing
15
or innovation. It would have been interesting to include more specific variables representing innovation or research
and development, but data availability prevented a more precise investigation of the effects of technical progress.
15
Arrow 1962.
The first results are presented in Table 1. Regression 1 shows that the coefficients on the lagged productivity and investment variables turn up with the
expected sign. Since b 0 is positive and inferior to 1, b is negative, so that a
conditional convergence effect is present. The coefficient for the education vari- able is neither significant nor has the expected sign. One notices that the impact of
the investment variable is quite strong. A difference of one percentage point in the investment to capital stock ratio gives a difference of over one percentage point in
the annual productivity growth rate.
Regression 2 adds the international specialisation variable. The coefficient for that variable is positive and significant. This implies that the mere fact of being
specialised in foreign trade is good for productivity growth, irrespective of the precise specialisation that the country has adopted. This contrasts with the results
found by Busson and Villa 1997 with cross-country regressions for 98 countries, where the specialisation variable had a significantly negative effect on growth.
Regression 3 includes the dissimilarity variable, which surprisingly appears with a significant, positive coefficient. This is again at odds with the results of Busson and
Villa, who found a significantly negative impact of trade dissimilarity in their estimations. One notices that the education variable now significantly contributes
negatively to productivity growth. This counter-intuitive result sometimes appear when one uses panel data instead of a cross-section of countries.
16
Basically, education levels have increased in all countries over the period considered, but
relatively more in some developed countries than in most developing countries. Therefore, this variable does not exhibit a convergence favouring pattern, which
explains its lack of significance or its negative contribution.
Regression 4 shows that a comparative advantage in electronics seems to have a positive effect on productivity growth. This complements the findings of regres-
sion 3. If specialisation matters, it does not imply that any specialisation is equivalent for growth. Regression 5 tests the inclusion of all three foreign trade
variables. The effects of the extra variables on the coefficients for lagged productiv- ity differentials, investment and education variables are minor. Both the specialisa-
tion and the comparative advantage in electronics variables turn up with positive and significant coefficients. Once the effect of specialisation is controlled for, a
comparative advantage in electronics appears as a positive influence on productivity growth. The trade dissimilarity variable fails to appear with a significant coefficient.
The seemingly negative impact of education on productivity growth is somewhat unsatisfactory. It runs against most growth theories as well as common sense. One
must nevertheless not forget that the sample of countries excludes most third world countries where one should expect the marginal impact of an increase in the
education level to be the largest. Among the restricted sample of countries chosen here, a rather crude indicator of education such as the one used so far may not be
enough to reflect the efficiency of the different national education and training systems. It is possible to exhibit a positive effect of education on productivity
16
See Berthe´lemy et al. 1996 on this issue.
growth when one takes an alternative indicator. Regressions in Table 2 include the percentage of the population that has completed secondary schooling as the
education variable. This time, the contribution of education is significantly positive in most cases, but has a somewhat limited impact on the growth of productivity.
The dissimilarity variable turns up with a significant positive sign. The positive influence of a specialisation in electronics is confirmed.
In order to assess more clearly whether the effect of education on productivity growth can be positive, the alternative variable the percentage of the population
that has completed secondary schooling is kept in the following regressions. Regression 1 in Table 2 shows that, when included alone, the effect of this
variable is positive but not significant. Attempts to include other education vari- ables higher education for instance were not met with more success. As in the
previous regressions, specialisation and a comparative advantage in electronics exert a positive influence on productivity growth.
It may be so that education has not only a direct effect on productivity growth, but an indirect effect as well. This indirect effect may travel through many channels,
but one is of particular interest here, the channel of international specialisation. A high level of schooling may exert positive or negative effects in interaction with a
certain pattern of international specialisation. For instance, a specialisation ori- ented towards electronics may be more or less beneficial to growth according to the
level of education of the labour force. This latter indicator would in fact partly reflect the ‘quality’ of the specialisation. in electronics, for instance, whether a high
comparative advantage simply reflects the domestic presence of many assembly lines of electronics goods. The nature of the electronics activities present in one
Table 2 Trade and growth
a
4 3
2 1
5 Dy
it−r
0.913 0.829
0.901 0.887
0.904 26.2
31.0 32.8
36.5 32.1
1.601 1.392
1.372 Dik
1.056 1.363
16.8 9.9
14.5 14.5
26.5 0.001
0.001 0.0002
Dedu 0.001
0.002 3.4
3.8 0.5
4.5 2.6
0.155 –
– Dspec.
0.246 –
– 7.2
– –
3.1 0.117
– Ldiss.
0.158 –
– –
– 3.9
– 2.5
DACelec –
– –
0.012 0.019
– –
– 3.9
6.6 0.246
M
2
0.857 0.509
0.322 0.665
29.7 26 29.7 29
Sargan test 28.3 28
29.4 27 26.9 28
117 117
Number of observations 117
117 117
a
Dependant variable, y
i−
y
it−,r
. T-statistics are given under the value of the estimated coefficients. The education variable is the percentage of the population that have completed secondary schooling.
Table 3 Interactions between trade variables and education
a
1 2
3 5
4 Dy
it−r
0.880 0.867
0.894 0.813
0.907 16.4
38.1 28.5
26.9 18.5
Dik 1.429
1.468 1.447
1.477 1.488
8.3 15.9
11.2 15.0
13.0 Dedu
– −
0.011 –
– −
0.022 –
– −
4.8 –
− 9.8
Dspec. –
0.151 0.168
– –
– 4.5
– 3.7
– D [edu·spec]
– –
– 0.006
− 0.004
– –
– 3.6
− 1.8
D [edu·diss] −
0.019 −
0.010 −
0.007 −
0.017 −
0.000 −
5.3 −
2.0 −
4.1 −
6.3 −
0.2 D [edu·ACelec]
0.003 0.004
0.002 0.005
0.002 5.5
8.0 5.4
6.0 10.9
M
2
− 0.147
0.381 0.447
0.632 −
0.454 Sargan test
22.9 22 23.6 27
28.7 27 27.5 25
24.9 27 Number of observations
117 117
117 117
117
a
Dependent variable, y
i−
y
it−r
.
country is expected to have influences on the type of positive spillovers associated with these activities. Low skill-intensive activities may reasonably be expected to
give less positive growth spillovers than high skill-intensive ones. Likewise, one may expect spillovers associated with electronics or another type of specialisation to be
all the more influential for productivity growth that the population or the labour force has a high level of schooling.
17
Therefore, it seems interesting to take into account explicitly the interactions between the level of education of the labour force
and international specialisation indicators. Regressions in Table 3 show the effects of interaction between education and
trade variables on productivity growth. A trade structure dissimilar to World trade structure is a negative factor for growth when associated with a high level of
schooling of the population. In fact, it is only in complementarity with a high level of schooling that trade dissimilarity appears to have negative effects on productivity
growth. Likewise, a higher level of education reinforces the positive effects of a trade specialisation oriented towards electronics. The specialisation variable the
Michaely index has a positive effect except in regression 5, when the education variable is taken out of the regression. It seems, therefore, difficult to assess the
precise effect of trade specialisation on productivity growth without specifying more precisely the pattern of specialisation.
17
This is related to the discussion concerning ‘institutional complementarity’ and the notion of ‘institutional comparative advantage’ Amable 2000.
5. Conclusion
